Keep vehicles off rooftops and bridge decks: probe the mesh where the vehicle is

On the photoreal stack the rendered-mesh floor took one probe at each coarse
cell's centre (0.001 degrees) and kept it for the session. A centre on a roof
passed the +80 m DEM window and lifted every vehicle in the cell to roof height
(a Haymarket bus drew 23.5 m above the street, on the garage roof); a centre in
the river put Red Line trains on the Longfellow Bridge 6-9 m under the deck.

The sampler now probes at the caller's point. A cell with a reading is probed
again only at spots about 17 m apart from earlier ones, eight at most, and a
spot refused only because the cell's DEM prior was cold is tried again later.
The floor keeps each reading that passes the gate at the point it was taken
(two within about 5 m keep the lower; sixteen at most per cell) and answers a
read from the reading nearest the asked point, so a viaduct and the street
beside it in one cell keep their own heights.

docs/CURRENT-STATE.md

@@ -1347,6 +1347,27 @@ and snapshot source. The standalone adapter keeps the existing USGS daily feed,
 M2.5+ filtering, static discs, magnitude labels and analyst records. Disabling or
 destroying the layer cancels pending work and ignores late results.
 
+## Rendered-mesh floor readings — November 9, 2026
+
+On the photoreal stack the shared floor (`cachedGroundFloor`) answers from
+rendered-mesh readings per coarse cell (0.001°, about 110 × 80 m), sampled by
+`src/services/meshFloorSampler.js` and gated against the DEM by
+`reportValidatedMeshFloorCell` in `src/services/groundFloor.js`. The probe goes
+where the caller's point is (a vehicle or contact), not to the cell's centre;
+the sampler reads a cell at up to eight spots about 12–17 m or more apart, and
+each reading that passes the gate is kept at its point (ground snaps add theirs
+the same way; a cell keeps at most sixteen). A read answers from the cell's
+reading nearest the asked point, so a viaduct and the street beside it in one
+cell keep their own heights; two readings within about 5 m of each other keep
+the lower. A spot refused only because the cell's DEM prior was cold is tried
+again later. Before this, one probe at the cell's centre latched for the
+session, so a centre on a roof lifted everything in the cell (a Haymarket bus
+drew 23.5 m above the street, on the garage roof) and a centre in the river put
+Red Line trains on the Longfellow Bridge 6–9 m under the deck. Residual: the
+probe is vertical, so a vehicle directly under a deck or a roof overhang reads
+the structure above it, and a vehicle between readings takes the nearest one, so
+its floor steps where the nearest reading changes.
+
 ## Vessel components and sources
 
 `src/data/aisLiveVessels.js` assembles `createVesselLayer` from the
@@ -3148,7 +3169,7 @@ This is the current runtime/source-of-truth snapshot for the project.
 >   `HELD_SNAP_CONTRADICTION_M` (5 m) ABOVE the held value drops the hold and the
 >   contact is COLD again, back to the floored 2D billboard. The rule is
 >   ONE-SIDED on purpose — it targets BURIAL. A cell reading BELOW a real sample
->   is the floor chain's expected under-read (one-shot latch over ~111 m,
+>   is the floor chain's expected under-read (a few point readings over ~111 m,
 >   neighbours lean lowest, `displayFloorHeightM` only ever raises), not evidence
 >   of anything, and a two-sided cut was disproved on the track rig by a planted
 >   cell 66.7 m below a real sample at the same spot. Accepted residual, stated